logo

SCIENCE CHINA Information Sciences, Volume 64 , Issue 9 : 192107(2021) https://doi.org/10.1007/s11432-020-2951-1

Large scale air pollution prediction with deep convolutional networks

More info
  • ReceivedFeb 17, 2020
  • AcceptedApr 15, 2020
  • PublishedAug 9, 2021

Abstract


Acknowledgment

This work was supported in part by National Science and Technology Major Project of the Ministry of Science and Technology of China (Grant No. 2018AAA0100701), National Natural Science Foundation of China (Grant Nos. 61906106, 62022048), and Institute for Guo Qiang of Tsinghua University and Beijing Academy of Artificial Intelligence.


References

[1] Stern A C. Air pollution: the effects of air pollution. Amsterdam: Elsevier, 1977. Google Scholar

[2] Brunekreef B, Holgate S T. Air pollution and health. Lancet, 2002, 360: 1233-1242 CrossRef Google Scholar

[3] Chow J C. Health Effects of Fine Particulate Air Pollution: Lines that Connect. J Air Waste Manage Association, 2006, 56: 707-708 CrossRef Google Scholar

[4] Dominici F, Peng R D, Bell M L. Fine Particulate Air Pollution and Hospital Admission for Cardiovascular and Respiratory Diseases. JAMA, 2006, 295: 1127-1134 CrossRef Google Scholar

[5] Yu-Fei Xing, Yue-Hua Xu, Min-Hua Shi, and Yi-Xin Lian. The impact of pm2. 5 on the human respiratory system. Journal of thoracic disease,2016, 8(1):E69, doi: 10.3978/j.issn.2072-1439.2016.01.19. Google Scholar

[6] Brook R D, Rajagopalan S, Pope Iii C A. Particulate Matter Air Pollution and Cardiovascular Disease. Circulation, 2010, 121: 2331-2378 CrossRef Google Scholar

[7] Pope III C A. Lung Cancer, Cardiopulmonary Mortality, and Long-term Exposure to Fine Particulate Air Pollution. JAMA, 2002, 287: 1132-1141 CrossRef Google Scholar

[8] James D E, Chambers J A, Kalma J D. Air quality prediction in urban and semi-urban regions with generalised input-Output analysis: The Hunter Region, Australia. Urban Ecol, 1985, 9: 25-44 CrossRef Google Scholar

[9] Bruckman L. Overview of the enhanced geocoded emissions modeling and projection (enhanced gemap) system. In: Proceeding of the Air & Waste Management Association's Regional Photochemical Measurements and Modeling Studies Conference, San Diego, 1993. Google Scholar

[10] Gu K, Qiao J, Li X. Highly Efficient Picture-Based Prediction of PM2.5 Concentration. IEEE Trans Ind Electron, 2019, 66: 3176-3184 CrossRef Google Scholar

[11] Corani G. Air quality prediction in Milan: feed-forward neural networks, pruned neural networks and lazy learning. Ecol Model, 2005, 185: 513-529 CrossRef Google Scholar

[12] Russo A, Raischel F, Lind P G. Air quality prediction using optimal neural networks with stochastic variables. Atmos Environ, 2013, 79: 822-830 CrossRef ADS arXiv Google Scholar

[13] Box G E P, Pierce D A. Distribution of Residual Autocorrelations in Autoregressive-Integrated Moving Average Time Series Models. J Am Statistical Association, 1970, 65: 1509-1526 CrossRef Google Scholar

[14] Urban air quality forecasting based on multi-dimensional collaborative Support Vector Regression (SVR): A case study of Beijing-Tianjin-Shijiazhuang. PLoS ONE, 2017, 12: e0179763 CrossRef ADS Google Scholar

[15] Zheng Y, Yi X W, Li M, et al. Forecasting fine-grained air quality based on big data. In: Proceedings of the 21th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, 2015. 2267--2276. Google Scholar

[16] Zheng Y, Liu F R, Hsieh H P. U-air: when urban air quality inference meets big data. In: Proceedings of the 19th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, 2013. 1436--1444. Google Scholar

[17] Kurt A, Oktay A B. Forecasting air pollutant indicator levels with geographic models 3days in advance using neural networks. Expert Syst Appl, 2010, 37: 7986-7992 CrossRef Google Scholar

[18] Ronneberger O, Fischer P, Brox T. U-net: convolutional networks for biomedical image segmentation. In: Proceedigns of International Conference on Medical Image Computing and Computer-Assisted Intervention, 2015. 234--241. Google Scholar

[19] He K M, Zhang X Y, Ren S Q, et al. Deep residual learning for image recognition. In: Proceedings of IEEE Conference on Computer Vision and Pattern Recognition, 2016. 770--778. Google Scholar

[20] Huang G, Liu Z, van der Maaten L, et al. Densely connected convolutional networks. In: Proceedings of IEEE Conference on Computer Vision and Pattern Recognition, 2017. 4700--4708. Google Scholar

[21] Kim Y, Fu J S, Miller T L. Improving ozone modeling in complex terrain at a fine grid resolution: Part I - examination of analysis nudging and all PBL schemes associated with LSMs in meteorological model. Atmos Environ, 2010, 44: 523-532 CrossRef ADS Google Scholar

[22] Baklanov A, Mestayer P G, Clappier A. Towards improving the simulation of meteorological fields in urban areas through updated/advanced surface fluxes description. Atmos Chem Phys, 2008, 8: 523-543 CrossRef ADS Google Scholar

[23] Jeong J I, Park R J, Woo J H. Source contributions to carbonaceous aerosol concentrations in Korea. Atmos Environ, 2011, 45: 1116-1125 CrossRef ADS Google Scholar

[24] Stern R, Builtjes P, Schaap M. A model inter-comparison study focussing on episodes with elevated PM10 concentrations. Atmos Environ, 2008, 42: 4567-4588 CrossRef ADS Google Scholar

[25] Li C, Hsu N C, Tsay S C. A study on the potential applications of satellite data in air quality monitoring and forecasting. Atmos Environ, 2011, 45: 3663-3675 CrossRef ADS Google Scholar

[26] Li X, Peng L, Yao X. Long short-term memory neural network for air pollutant concentration predictions: Method development and evaluation. Environ Pollution, 2017, 231: 997-1004 CrossRef Google Scholar

[27] Gu K, Qiao J, Lin W. Recurrent Air Quality Predictor Based on Meteorology- and Pollution-Related Factors. IEEE Trans Ind Inf, 2018, 14: 3946-3955 CrossRef Google Scholar

[28] Gu K, Xia Z, Qiao J. IEEE Trans Instrum Meas, 2020, 69: 660-671 CrossRef Google Scholar

[29] Yi X W, Zhang J B, Wang Z Y, et al. Deep distributed fusion network for air quality prediction. In: Proceedings of the 24th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining, 2018. 965--973. Google Scholar

[30] Zheng Y, Capra L, Wolfson O. Urban Computing. ACM Trans Intell Syst Technol, 2014, 5: 1-55 CrossRef Google Scholar

[31] Xiong Z, Sheng H, Rong W G. Intelligent transportation systems for smart cities: a progress review. Sci China Inf Sci, 2012, 55: 2908-2914 CrossRef Google Scholar

[32] Deng M, Liu Q L, Wang J Q. A general method of spatio-temporal clustering analysis. Sci China Inf Sci, 2013, 56: 1-14 CrossRef Google Scholar

[33] Wang C M, Hu X P, Yao L. Spatio-temporal pattern analysis of single-trial EEG signals recorded during visual object recognition. Sci China Inf Sci, 2011, 54: 2499-2507 CrossRef Google Scholar

[34] Wang W, Hu C B, Chen N C. Spatio-temporal enabled urban decision-making process modeling and visualization under the cyber-physical environment. Sci China Inf Sci, 2015, 58: 1-17 CrossRef Google Scholar

[35] Lv Y, Duan Y, Kang W. Traffic Flow Prediction With Big Data: A Deep Learning Approach. IEEE Trans Intell Transp Syst, 2014, : 1-9 CrossRef Google Scholar

[36] Zhang J B, Zheng Y, Qi D K. Deep spatio-temporal residual networks for citywide crowd flows prediction. In: Proceedings of the 31st AAAI Conference on Artificial Intelligence, 2017. Google Scholar

[37] Liang Y X, Ke S Y, Zhang J B, et al. Geoman: multi-level attention networks for geo-sensory time series prediction. In: Proceedings of the 27th International Joint Conference on Artificial Intelligence, 2018. 3428--3434. Google Scholar

[38] Zhang J B, Zheng Y, Qi D K, et al. Dnn-based prediction model for spatio-temporal data. In: Proceedings of the 24th ACM SIGSPATIAL International Conference on Advances in Geographic Information Systems, 2016. Google Scholar

[39] Parzen E. On Estimation of a Probability Density Function and Mode. Ann Math Statist, 1962, 33: 1065-1076 CrossRef Google Scholar

[40] Ioffe S, Szegedy C. Batch normalization: accelerating deep network training by reducing internal covariate shift. 2015,. arXiv Google Scholar

[41] Nair V, Hinton G E. Rectified linear units improve restricted boltzmann machines. In: Proceedings of the 27th International Conference on Machine Learning (ICML-10), 2010. 807--814. Google Scholar

[42] Kingma D P, Ba J. Adam: a method for stochastic optimization. 2014,. arXiv Google Scholar

[43] Broomhead D S, Lowe D. Radial Basis Functions, Multi-Variable Functional Interpolation and Adaptive Networks. Royal Signals and Radar Establishment Malvern (United Kingdom) Technical report, 1988. Google Scholar

[44] Friedman J H. Greedy function approximation: a gradient boosting machine. Ann Stat, 2001, 29: 1189--1232. Google Scholar

[45] Hochreiter S, Schmidhuber J. Long Short-Term Memory. Neural Computation, 1997, 9: 1735-1780 CrossRef Google Scholar

  • Figure 1

    (Color online) A comparison of the number of monitoring stations in previous work [16](a) and this paper (b). Both sub-figures show a region of approximately 70 km $\times$ 70 km. The density of observation points is about ten times higher than that in previous work.

  • Figure 2

    (Color online) An illustration of the transformation of our observed data. The observed PM2.5 concentrations of all monitoring points at time $t_1$ are firstly transformed into a pollution map according to the stations' geolocation. Then the pollution map is imputed into a regular image, and this transformation is conducted for all observed hours to consist of the input tensor to facilitates the use of a deep convolutional network.

  • Figure 3

    (Color online) The architecture of the proposed convolutional network.

  • Figure 4

    (Color online) A comparison of the proposed model, temporal model and existing spatiotemporal model for AQP at two randomly selected monitoring stations. (a) Monitor 10965's result; (b) monitor 16038's result.

  • Figure 5

    (Color online) The prediction of the proposed method (a), the real observation values (b), and mean absolute error (c) at a randomly selected time slot.

  • Table 1  

    Table 1Dataset description

    Item Description
    Monitoring points 393
    Collection time periodSeptember/1st/2018–November/30th/2018
    May/1st/2019–June/30th/2019
    Resolution 1 hour per record
    PM2.5 concentrations $\mu$g/m$^3$
    Longitude Location information
    Latitude Location information
  • Table 2  

    Table 2One-hour AQP results

    Type Model MAE MAPE Time (s)
    Temporal GBDT 23.81 79.61 0.0490
    SVR 20.96 81.47 0.0184
    LSTM 16.84 83.28 0.1190
    1D CNN 17.58 82.15 0.1231
    Spatiot- PA 17.89 82.36 0.0223
    emporal Our model 13.26 85.04 0.0070
  • Table 3  

    Table 3Three-hour AQP results

    Our methodPA
    Hour MAEMAPEMAEMAPE
    The 1st 15.38 83.37 17.89 82.36
    The 2nd 19.36 77.52 26.91 72.60
    The 3rd 25.24 71.17 34.66 64.29
  • Table 4  

    Table 4Performance v.s. the number of monitors points

    Monitors pointsMAEMAPE
    20 21.54 72.21
    30 20.54 72.98
    50 14.12 78.68
    393 13.26 85.04
  • Table 5  

    Table 5Study of the effect of incorporating spatial information

    Model MAEMAPE
    Spatial information 13.26 85.04
    Non-Spatial information 15.04 83.32
  • Table 6  

    Table 6Study on the effectiveness of the skip connection

    Model MAEMAPE
    Our method 13.26 85.04
    Our method-nonSkip 14.74 84.29
  • Table 7  

    Table 7Study on the effectiveness of the dense connections

    Model MAEMAPE
    Our method 13.26 85.04
    Without-dense-connections 15.0083.36
qqqq

Contact and support